Review - Agenda 2020

Many companies have tried to apply AI into their business. Yet, 75% are not seeing any value yet. Based on a global survey and interviews he looked at what it takes to become successful with AI. What do pioneers do right, what should you avoid.

Modern Enterprise is a complex, often siloed ecosystem with cross departmental dependencies. AI and advanced analytics is recently making an impact across industries and various job functions from product development, through manufacturing, supply chain to sales and marketing. This talk will be divided into two main parts. Marcin will start with two real life case studies demonstrating in detail how modern predictive analytics methods can be applied to product creations (Yes, AI can automate some of the hard parts of the product creation in many industries, e.g. pharma or chemicals) and to manufacturing optimisation. In the case of manufacturing production line optimization he will show what methodology was applied to help a Swiss chemical producer achieve 8-15% OEE (Overall Equipment Effectiveness) improvements by leveraging advanced analytics and data science.  In the second part, based on extensive Unit8 experience in the industry Marcin will take a step back and summarise key learnings and best practises that lead to AI & Analytics projects delivery with positive ROI in Industry 4.0 sector.

We are aware, Digital Twin concept helps monitor the product and its performance while in use. Here is an effort to bring digital twin concept for vehicles specifically meant for farming. Tractors have accessories attached to them to perform tasks like, cutting, pesticide spraying, seeding etc., in the fields. An Android Application automatically turns-on and records the sound, when in field. An ensemble of Machine learning algorithms assess the sound and identify the tasks performed in the field, “real time”. The approach provides 95% accuracy in identifying the tasks on true data. The approach is different from the normal in that it takes only one data type and source – sound waves (compared to multiple sensors data, vehicle driving details etc.) and is able to analyse multiple aspects.

ZF plant Saarbrücken manufactures around 11,000 transmissions per day with a large number of variants. An AI project was started to get reliable and fast results on root cause discovery. Speed is important because production runs 24 / 7. The target is to reduce waste in certain manufacturing domains by 20%. Key success factor is fast detection within the production chain by AI. Complex root-cause findings can be reduced from several days to hours. Self-learning AI solution Predictive Intelligence from IS Predict was used to analyze complex data from production, assembly + quality and to realize transparency on disturbing factors.

How can Data Science help companies to optimize processes and perhaps even open up new business areas and services? How do analytical prototypes develop into Mission Critical Services? Find out how eoda has helped TRUMPF, the technology and world market leader in lasers and laser systems, to reduce downtimes, ensure the high level of machine manufacturing and develop new business models in the process. With Deep Learning, recurring patterns in machine and sensor data can be identified. In this way, self-optimizing algorithms can be developed and an intelligent predictive maintenance system was realized. With this Condition Monitoring Portal, which evolved into a stand-alone data science platform, it is possible to detect errors and anomalies in advance and thus prevent unforeseeable machine failures, as well as to optimize processes globally and increase machine availability.

Clariant is a specialty chemicals company that founded Navigance GmbH in order to venture into their first digital product, a solution that optimizes the efficiency of the chemical process. 

As safety of the operation takes highest priority, fast acceptance for such new technology can only be won without full automation. Therefore it is crucial that the staff stays in the loop and continues to take the responsibility of changes in system. However, the product team learned about a big challenge regarding the human-machine interaction: Process engineers have an already demanding work routine, so any distraction like another crowded monitor with cryptic alerts was not wanted and would not have been accepted. In addition, it was not easy to gain their trust on recommendations from the black box model. Without a good way to address this, they could easily feel left behind as experts. Evelyn Münster and Sebastian Werner will show you how they developed purpose-built data visualizations that help gaining trust in the machine learning black box. They will talk about how they designed a user interface that leaves the users in control and caters to their workflow, ensuring high user acceptance.

The equipment used for plastic paper production is highly complex. Up to 5 industrial controllers control the processes and record data from up to 400 sensors (e.g. temperatures, pressures, speeds, etc.) to be used to detect problems at an early stage and avoid production downtimes.

The sensor data was read into MATLAB® and prepared for the use of ML algorithms. From a series of different algorithms, the one with the best training results (bagged decision trees) was finally selected and applied to the machine data in production. In addition, the algorithm was integrated into the existing IT infrastructure and indicates to the operator via a user interface whether an intervention is necessary.

By using AI methods for predictive maintenance, significant savings can be achieved by avoiding production losses and machine downtime. The integration into the existing IT system allows the application to run 24/7 and supports the operating personnel in ensuring the expected production quality.

Opportunities to apply ML to solve business challenges abound throughout Industry. To really move the needle, only the highest-value opportunities should be chosen. Erika from Dow, Inc. will present a case study on how ML has been leveraged to solve an array of business challenges, ranging from forecasting to sentiment analysis, or classification to neural networks, leading to successful outcomes across the enterprise.

The mission of the Swiss Data Science Center is to accelerate the adoption of data science and machine learning techniques within the academic community and the industrial sector. Throughout the projects, they found out that companies encounter similar challenges, no matter their industry sector. The digital transformation journey is a marathon which needs the right mindset, agility and skills. In this talk, Sandro will highlight lessons learnt from working with companies in their effort to become more data-driven. He will discuss topics such as selecting relevant use cases, getting the right data, collaborating with domain experts and monitoring projects with canvas. He will discuss solutions to data and humans’ challenges, as well as best practices to facilitate this journey. His presentation will be based on experience gathered with leading companies in sectors such as manufacturing, banking and pharma.

Why are data science projects in an industrial context still rare? Simon made the experience that, especially in complex industry facilities, understanding the data and identifying a concrete use case requires a lot of domain expertise. Building up this domain expertise might be the hardest part for a data scientist. But without clear data understanding there won’t be a precise problem definition and sooner or later the project will fail. In this talk, Simon will explain the key success factors of a concrete data science project with TAL-Group, which is the operator of the transalpine oil pipeline transporting about 45 millions of tons of crude oil every year across the alps. Based on the experiences made in that project he will focus on two questions: 1. How did they overcome this “hardest part” from raw sensor data and to a concrete data science problem definition? 2. How did machine learning algorithms help in order to identify root causes for efficiency losses while pumping raw oil?

For railway operators like the National Express Rail GmbH, ordering the correct amount of energy in advance will reduce costs significantly. However, predicting power consumption is difficult since consumption varies due to delays of trains, holidays, weather events and unscheduled rides. By incorporating internal and external data sources on these subjects, we developed a machine learning solution to predict the energy needs of the National Express train fleet. 

The talk will illustrate the challenges of choosing the correct machine learning techniques for this specific application. Furthermore, we will discuss how risk evaluation plays a crucial role in the business evaluation of our forecast.

Important information is often buried in graphical plans, hand written notes and other sources which are almost impossible to access at scale, even with the help of computer vision or optical character recognition. With Active Learning, you get a huge reduction in human effort in return for a somewhat lower precision. Sarah and Michael demonstrate how they successfully applied Active Learning to an archive consisting of partly handwritten plans which can only be interpreted with expert knowledge on company specifics. They discuss which algorithms from deep learning and classical ensemble learning can be used together with this technique.

Munich Airport is one of the most important hubs in Europe and is again awarded as the best airport in 2019 like several years before. To achieve a comfortable passenger journey you have to provide excellent operational processes. To make this also possible in the future, predictive analytics helps to identify bottlenecks and calculates future scenarios. The case study by Dr. Heike Markus, Munich Airport, provides insights in the development of predictive models to quantify the impact of strategic decisions on major investment projects. She will talk about the challenges of implementing new data-driven work routines and the increasing requirements of flexibility and transparency of business models. She will explain what data analytics can achieve and what difficulties have to be overcome to be successful.

Judging by Amazon’s success, the recommender system works. Adopting such a system to the automotive aftermarket poses many challenges. Association analysis for vehicles is much more complicated than for simple consumer goods. Normally a recommender system is tailored towards the users’ subjective preferences whereas in the aftermarket the user follows an objective search approach. TecAlliance implemented a recommender system in its spare parts catalogue – known as TecDoc Catalogue – based on a simple association analysis and web analytics data that overcomes this challenge. Workshops don’t have to search interrelated articles manually anymore and get a suggestion of similar articles that are related to the viewed one. Overall, this leads to a massive time saving and better process design.